Material property prediction device for rolled product

Abstract

This material property prediction device for a rolled product comprises: an approximation model generation unit for generating, off-line, an approximation model by which to comprehensively predict material properties of group of rolled products produced by a rolling line; and a material property prediction unit for predicting, by using the approximation model, material properties in each three-dimensional mesh-like region of the rolled products produced by the rolling line. The approximation model generation unit comprises: a data set generation unit which has a condition setting part for setting rolling conditions for the group of rolled products and a material calculation part for calculating material properties and metallurgical phenomena under the rolling conditions and which generates a data set used for generating the approximation model; and a model parameter determination unit for determining parameters for representing the approximation model by using the data set.

Description

Rolled product material property prediction device

[0001] The present disclosure relates to a device for predicting material properties of a rolled product, and more particularly to a device for predicting material properties of a rolled product produced in a hot rolling process.

[0002] The material properties of rolled products (hereinafter also referred to as "products") made of metal materials such as steel vary depending on their alloy composition and the heating, processing, and cooling conditions of the hot rolling process. Examples of material properties include mechanical properties (strength, formability, toughness, etc.) and electromagnetic properties (magnetic permeability, etc.). The alloy composition is adjusted by controlling the amount of added component elements. During this composition adjustment, a single lot is large, for example, using a composition adjustment furnace capable of holding approximately 100 tons of molten steel. Therefore, it is impossible to change the amount of added elements for each individual rolled product, which typically weighs approximately 15 tons. Therefore, in order to manufacture hot-rolled product coils with the desired material properties, it is important to appropriately control the heating, processing, and cooling conditions.

[0003] In a hot rolling process, different rolled products are produced by changing the target values ​​of various process parameters, which are process conditions related to product quality and operating conditions. The process parameters include, for example, target temperatures at each point on the rolling line, such as the entry and exit temperatures of the finishing mill and the coiling temperature, schedules related to thickness and width reduction, such as the transfer bar thickness at the exit of the roughing mill and the reduction rate of each pass, whether or not descalers provided in the finishing mill and roughing mill are used for each pass, whether or not interstand cooling disposed between the stands of the roughing mill and the finishing mill is used and the initial flow rate, the amount of lubricating oil used in the finishing mill, and the cooling pattern used in the run-out table.

[0004] Conventionally, process parameters related to heating, processing, and cooling, such as target values ​​for heating temperature, target values ​​for dimensions after processing, and target values ​​for cooling rate, are set for each specification of a rolled product, and a method of controlling temperature and dimensions to achieve these target values ​​has been generally adopted. Note that while the target values ​​for product dimensions are specified in advance, the target values ​​for thickness, temperature, and cooling rate at the outlet of each stand have been determined based on many years of experience. However, in recent years, requirements for product specifications have become significantly more sophisticated and diversified, and there are cases where these target values ​​cannot necessarily be determined appropriately using methods based on experience.

[0005] Patent Document 1 listed below discloses an apparatus for offline simulation of a manufacturing process using a process model that models each of the manufacturing processes of heating, processing, and cooling, in order to examine in advance whether a product manufactured under a certain alloy composition and process parameters will have the desired product quality.

[0006] Furthermore, there is a growing need for stricter management of material properties than the traditional guaranteed range. Conventionally, as specified in the Japanese Industrial Standards (JIS), the condition (tolerance) was that material properties exceed a standard value. For example, a tensile test was performed on a sample taken from a product to determine whether the measured value exceeded the standard value. However, in recent years, higher precision has been required even in post-production processes. The above-mentioned conventional tolerance ranges may not be sufficient for downstream forming processes (drawing, bending, pressing, etc.). Examples of problems include the rolled material being too hard to form, the springback (elastic recovery) after pressing being too large, resulting in poor shape fixability, or the edge cracking during forming. Therefore, empirical setting methods and material property management methods have not always been able to adequately control the above-mentioned target values.

[0007] A conventional method for managing the rolling process uses the output values ​​of thermometers installed in the rolling line to manage the temperature of the entire rolled coil and also manage material properties that are closely related to the rolling temperature. Specifically, thermometers are installed at the exit of the heating furnace, the entry and exit of the roughing mill, the entry and exit of the finishing mill, the entry of the coiler, etc., of the rolling line. The thermometers measure the temperature of the central portion of the rolled material in the width direction (hereinafter simply referred to as the "width direction"). Then, a host computer controls the output values ​​from the thermometers so that they coincide with a target temperature determined based on experience. Thus, in the past, the material properties of the rolled material in the width direction were not taken into consideration when managing the rolling process.

[0008] The widthwise ends (edges) of the material being rolled tend to cool easily, resulting in a temperature difference between them and the center. A rolling line may be equipped with devices to raise the temperature of the widthwise ends of the material being rolled or to prevent a drop in the temperature of the widthwise ends of the material being rolled. For example, an edge mask is used to prevent cooling water from splashing on the edge during cooling after rolling. Furthermore, before finish rolling, the edge is heated using an induction heating device such as an edge heater.

[0009] Furthermore, in recent years, scan pyrometers have been installed before and after the device in some cases to verify the effectiveness of the device. Scan pyrometers can be used to measure the temperature distribution in the width direction of the rolled material. Furthermore, scan pyrometers are also used in multi-gauges, which have been adopted in rolling lines in recent years, to use the temperature distribution in the width direction of the rolled material to correct the measured values. Multi-gauges are composite measuring instruments that measure thickness, crown, width, etc., all in one unit, and their measurement accuracy has improved significantly in recent years.

[0010] Thus, equipment for measuring the temperature distribution in the width direction of the rolled material is being introduced into rolling lines. Furthermore, some attempts have been made to calculate the temperature distribution in the width direction of the rolled material and use this for control. Patent Document 2 listed below discloses a means for calculating the temperature distribution in the thickness direction and width direction of the rolled material. Patent Document 3 listed below also discloses a method for uniformizing the temperature in the width direction by controlling edge heaters based on a calculation of the temperature distribution in the width direction.

[0011] In addition, when unsatisfied portions occur in terms of material quality, as with thickness, width, and shape, they are generally required to be trimmed off at the dividing line in the downstream process, taking into account margins and comparing actual data. Although the amount to be trimmed off is directly related to yield, in the past, the amount to be trimmed off was only roughly determined. In light of this situation, it has been desirable to optimize the amount to be trimmed off, even when unsatisfied portions occur, from the perspective of improving yield, and further to minimize unsatisfied portions through process improvements.

[0012] For this reason, conventionally, attempts have been made to control the properties by predicting the properties of the rolled product using a property prediction model, using manufacturing conditions such as heating, processing, and cooling in the rolling process as inputs.

[0013] Patent Document 4 listed below discloses a method for model learning using actual values ​​of mechanical properties obtained from mechanical property measurement tests such as tensile tests and structure observations conducted on some product coils, for a model that mathematically expresses metallurgical phenomena to predict changes in the microstructure of a rolled material and the mechanical properties of the final product. Patent Document 5 listed below discloses a method for outputting a material distribution that associates positions in two dimensions, namely the longitudinal and transverse directions, with material characteristic values. Patent Document 6 listed below discloses a method for accumulating operating conditions and material performance data, searching for similar operating conditions, and estimating the material of every position on a product coil in a mesh-like manner online. Furthermore, Patent Document 7 listed below discloses a method for predicting material properties using a neural network.

[0014] Japanese Patent No. 6292309 Japanese Patent No. 6197676 Japanese Patent No. 6447710 Japanese Patent No. 5396889 Japanese Patent Publication No. 2022-48037 Japanese Patent No. 6086155 Japanese Patent Publication No. 2005-315703

[0015] However, material property predictions based on mathematical models of metallurgical phenomena, such as those shown in Patent Documents 4 and 5, impose a heavy computational load on highly accurate models that faithfully model micro-phenomena. For example, even if there are only a few calculation points in a product coil, it can require several seconds of calculation time. For example, if a 1-km product coil is to be predicted with a mesh coarseness of 2 m pitch in the rolling direction, 3 points in the thickness direction, and 5 points in the width direction, the number of calculation points will be 7,500, and assuming a calculation time of 1 second per calculation point, it will require more than two hours of calculation time. Rolling operations often involve rolling the same or similar product categories consecutively. If the results of the material are to be reflected in operational changes for the next material or materials within the same lot, this takes too long, even though one coil may be rolled every 2 to 5 minutes.

[0016] Furthermore, in Patent Documents 6 and 7, the relationship between past operational conditions and material performance is explored and modeled, and the material properties of newly rolled product coils are predicted based on empirical rules. Such empirical models are generally known to be fast and contribute to solving the problem of computational load. However, models based on past operational conditions and material performance have limitations in approximation accuracy when modeling complex material behavior, and sufficient prediction accuracy cannot be expected.

[0017] The present disclosure has been made to solve the above-mentioned problems, and an object of the present disclosure is to provide a rolled product material property prediction device that can quickly and accurately predict the material property of an entire group of rolled products online.

[0018] A first aspect relates to a rolled product material property prediction device for predicting material properties of rolled products manufactured on a rolling line. The rolled product material property prediction device comprises an approximation model creation unit that creates an approximation model offline that comprehensively predicts material properties of a group of rolled products manufactured on the rolling line, and a material property prediction unit that predicts material properties in each three-dimensional mesh-like region of the rolled products manufactured on the rolling line online using the approximation model created by the approximation model creation unit. The approximation model creation unit has a condition setting unit that sets rolling conditions for the group of rolled products, and a material calculation unit that calculates metallurgical phenomena and material properties under the rolling conditions, and also comprises a dataset creation unit that creates a dataset used to create the approximation model, and a model parameter determination unit that uses the dataset to determine parameters that represent the approximation model.

[0019] The second aspect has the same features as the first aspect, but further includes the following: the data set creation unit creates the data set using the rolling conditions set by the condition setting unit as explanatory variables and the material properties calculated by the material calculation unit as objective variables.

[0020] The third aspect has the following features in addition to the features of the first aspect: the material property prediction unit comprises: a rolling data collection unit that collects online rolling data obtained when a rolled product is manufactured on the rolling line, a model input creation unit that creates online input data for the approximation model from the rolling data collected by the rolling data collection unit, an approximation model calculation unit that calculates online material properties of each zone in a three-dimensional mesh of a product coil by inputting the input data created by the model input creation unit into the approximation model, and a material property output unit that outputs the material properties of each zone calculated by the approximation model calculation unit, information representing the position of each zone in the rolled product, and information related to the material properties.

[0021] The fourth aspect has the same features as the first aspect, but further includes the following: the approximation model is a machine learning model.

[0022] A fifth aspect has the following feature in addition to any one of the first to fourth aspects: the material property prediction unit includes a material property correction unit that corrects the material property calculated by the approximation model calculation unit using the approximation model, using a material property result calculated using a metallurgical phenomenon model that mathematically expresses a metallurgical phenomenon.

[0023] According to the present disclosure, by creating in advance data sets corresponding to rolling conditions (temperature conditions, processing conditions, time / speed conditions) that have not been implemented in an actual rolling process or rolling conditions that have little experience of implementation, and by creating an approximation model offline using the created data sets, it is possible to create an approximation model that can be used for the entire group of rolled products and has high approximation accuracy. By using the approximation model created in this way, it is possible to reduce the calculation load for online prediction of material properties in each zone of a three-dimensional mesh of a rolled product. Moreover, because it is possible to quickly calculate material properties of every zone (part) of a rolled product, it is also possible to reflect the results of the rolled product in operational changes for the next rolled product or for rolled products in the same lot.

[0024] 1 is a diagram showing an example of a hot sheet rolling line to which the material property prediction device for rolled products according to embodiment 1 is applied. FIG. 2 is a block diagram showing a rolling system according to embodiment 1. FIG. 3 is a block diagram showing the functions of the material property prediction device for rolled products according to embodiment 1. FIG. 4 is a diagram showing an example of the hardware configuration of the material property prediction device for rolled products. FIG. 5 is a block diagram showing the configuration of an approximation model creation unit. FIG. 6 is a diagram for explaining a dataset created by the dataset creation unit. FIG. 7 is a schematic diagram for explaining an example of a cooling pattern for explanatory variables. FIG. 8 is a diagram for explaining the configuration of the dataset creation unit and the creation of a dataset. FIG. 9 is a diagram for explaining the setting of the material calculation unit and explanatory variables by the condition setting unit. FIG. 10 is a diagram for explaining metallurgical phenomenon calculation and setting of target variables by the material calculation unit. FIG. 11 is a schematic diagram showing an example of an approximate model. FIG. 12 is a block diagram showing the configuration of a material property prediction unit. FIG. 13 is a schematic diagram showing an example of a mesh area for which the approximation model calculation unit calculates material properties. FIG. 14 is a schematic diagram showing an example of output from a material property output unit. FIG. 15 is a block diagram showing the configuration of a material property prediction unit in embodiment 2. FIG. 16 is a schematic diagram showing a correction method in the material property correction unit in embodiment 2.

[0025] Hereinafter, with reference to the drawings, an embodiment of the present invention will be described in detail, taking as an example the case of predicting the material properties of a rolled product manufactured in a hot rolling line. Elements common to the various drawings will be assigned the same reference numerals, and duplicated explanations will be omitted.

[0026] Embodiment 1. [Rolling Line] Fig. 1 is a diagram showing an example of a hot sheet rolling line (hereinafter also referred to as a "rolling line") to which a material property prediction device for a rolled product (hereinafter also referred to as a "prediction device") according to embodiment 1 is applied. In this embodiment, a prediction device that predicts the material property of a rolled product manufactured in the rolling line shown in Fig. 1 will be described, but the prediction device of the present disclosure can also be applied to other rolling lines.

[0027] The rolling line includes a heating device, a rolling mill, a cooling device, a downcoiler, and a conveying table connecting these devices. These devices are driven by actuators such as electric motors and hydraulic devices. Specifically, the rolling line 1 shown in Fig. 1 includes, in order from the upstream side of the conveying table, a heating furnace 2, a high-pressure descaling device 3, a roughing entry thermometer 4, a roughing edger 5, a roughing horizontal rolling mill (hereinafter also referred to as the "roughing mill") 6, a roughing exit thermometer 7, an edge heater 8, a bar heater 9, a finish entry thermometer 10, a crop shear 11, a finish entry descaling device 12, an F1 edger 13, a finish rolling mill 14, a multi-gauge 15, a finish exit thermometer 16, a run-out table 17, a coiler entry thermometer 18, and a downcoiler 19.

[0028] The heating furnace 2 is a furnace for heating the material to be rolled (slab) and is controlled to obtain the desired slab heating pattern and heating furnace exit temperature. In the following description, the material to be rolled includes not only slabs and steel plates, but also intermediate states up to the completion of a product coil. The high-pressure descaling device 3 sprays high-pressure water from above and below the material to be rolled after it leaves the heating furnace 2 to remove scale from the surface of the material. The roughing entry thermometer 4 is located on the entry side (upstream) of the roughing mill 6 and measures the roughing entry temperature, which is the temperature of the surface (e.g., the top surface) at the center of the width direction of the material to be rolled. The roughing edger 5 rolls the material to be rolled in the plate width direction. The roughing mill 6 rough rolls the steel plate in the plate thickness direction. The roughing mill 6 rolls the material to be rolled in multiple passes to achieve the desired thickness. For this reason, a reversing rolling mill can be used as the roughing mill 6. The rough-outlet-side thermometer 7 measures the temperature of the surface (e.g., the upper surface) of the material to be rolled. The rough-outlet-side thermometer 5 is arranged on the outlet side (downstream side) of the roughing mill 6. When the material to be rolled passes through the roughing mill 6, the rough-outlet-side thermometer 7 measures the surface temperature of the central part in the width direction as the rough-outlet-side temperature.

[0029] The edge heater 8 is a device that heats the widthwise edge of the rolled material by electromagnetic induction heating or the like to control the temperature of the rolled material. The bar heater 9 is a device that heats the entire rolled material by electromagnetic induction heating or the like to control the temperature of the rolled material. The finish entry thermometer 10 is located on the entry side of the finishing rolling mill 14 and measures the finish entry temperature, which is the temperature of the surface (e.g., the top surface) of the central portion of the widthwise center of the rolled material. The crop shear 11 cuts off the leading and trailing ends of the steel plate. The finish entry descaling device 12 removes scale from the surface of the steel plate at the entry side of the finishing rolling mill 14. The F1 edger 13 is located on the entry side of the finishing rolling mill 14 and its rollers contact the rolled material from the side. The F1 edger 13 deforms the rolled material to narrow its width without buckling. The finishing mill 14 is made up of one or more stands, and is a tandem finishing mill made up of seven stands in the example shown in Fig. 1. The finishing mill 14 finish-rolls the material to be rolled to a predetermined plate thickness.

[0030] The multi-gauge 15 is a composite measuring instrument capable of performing various measurements with a single device. The multi-gauge 15 has, for example, a configuration in which multiple X-ray detectors are arranged in the width direction of the rolled material. The multi-gauge 15 measures, for example, the thickness distribution in the width direction of the rolled material. One multi-gauge 15 can measure the thickness, crown, and width of the rolled material. In recent years, the measurement accuracy of the multi-gauge 15 has significantly improved. For this reason, it is more cost-effective to purchase one multi-gauge 15 than to purchase a thickness gauge, crown gauge, and width gauge separately, and the introduction of multi-gauges 15 into hot rolling lines is progressing. The multi-gauge 15 is equipped with an internal thermometer and a scan pyrometer. The multi-gauge 15 measures the temperature of the rolled material and uses the measurement value to correct the detection value of the X-ray detector.

[0031] The finish exit thermometer 16 measures the temperature of the surface (e.g., the upper surface) of the rolled material. The finish exit thermometer 16 is arranged on the exit side (downstream side) of the finish rolling mill 14. The finish exit thermometer 16 measures the surface temperature of the center part in the width direction of the rolled material that has passed through the finish rolling mill 14 as the finish exit temperature. The finish exit thermometer 12 is arranged on the exit side of the finish rolling mill 10. The finish exit temperature of the rolled material is closely related to the formation of the metal structure and material properties (tensile strength, yield stress, elongation, etc.) of the product. For this reason, the finish exit temperature of the rolled material needs to be properly managed.

[0032] The run-out table 17 is a cooling device that cools the material to be rolled with cooling water in order to control the temperature of the rolled product. The run-out table 17 supplies cooling water from nozzles to the surface of the material to be rolled, for example, to control the temperature of the material. The run-out table 17 is equipped with a large number of nozzles in the longitudinal direction of the material to be rolled (the conveying direction of the conveying table). These nozzles are divided into multiple banks. Nozzle control is performed for each bank to control the cooling rate of the material to be rolled. Water cooling is performed in the bank that supplies cooling water, and air cooling is performed in the bank that does not supply cooling water. The rolling line may further be equipped with a cooling device such as a cooling table or a forced cooling device.

[0033] The coiler entry thermometer 18 is arranged on the entry side (upstream side) of the down coiler 19. After the material to be rolled passes through the run-out table 17, the coiler entry thermometer 18 measures the surface temperature of the central part in the width direction as the coiling temperature. The finish exit thermometer 12 is arranged on the exit side of the finish rolling mill 10. The coiling temperature of the material to be rolled is closely related to the formation of the metallographic structure and material properties (tensile strength, yield stress, elongation, etc.) of the product. For this reason, the coiling temperature of the material to be rolled needs to be appropriately controlled.

[0034] The down coiler 19 is a device for winding the rolled product and shaping it into a shape that is easy to transport. The transport table is a device for transporting the rolled product from each process to the next process. These devices are driven by actuators such as electric motors and hydraulic devices.

[0035] The rolling line 1 shown in FIG. 1 further includes a scan pyrometer 20. The scan pyrometer 20 measures the temperature of the surface (e.g., the top surface, or the top and bottom surfaces) of the rolled material at least at multiple locations in the width direction of the rolled material. The scan pyrometer 20 is preferably disposed before and after a device for improving the temperature of the rolled material. In the example shown in FIG. 1, the scan pyrometers 20 are installed in front of the edge heater 8, behind the bar heater 9, and before and after the run-out table 17. The scan pyrometer 20 disposed on the inlet side of the run-out table 17 is provided inside the multi-gauge 15.

[0036] [Rolling System] Figure 2 is a block diagram showing a rolling system 21 according to embodiment 1. The rolling system 21 is a control system for the rolling line 1, and has a hierarchical structure from level 0 to level 3. Level 0 has a drive control device that controls the electric motors that drive each device in the rolling line 1, and hydraulic equipment (hydraulic devices) that drive each device in the rolling line 1. Level 1 has a control controller 24. Level 2 has a setting computer 23. Level 3 has a host computer 22 for production management. A material property prediction device 25 of a rolled product, which will be described later, is connected to the setting computer 23 and is able to receive rolling data.

[0037] In the hot rolling process, different products are produced by changing the process conditions related to product quality and operating conditions, i.e., the target values ​​of various process parameters. The process is controlled by the setting computer 23 so as to achieve the target product quality, i.e., so as to achieve the target values ​​of the various process parameters.

[0038] The target values ​​of the process parameters may be specified by a level 3 host computer 22 that is higher than the level 2 setting computer 23. Alternatively, the target values ​​of the process parameters may have a table in a database belonging to the setting computer 23 and may be specified using the steel grade, plate thickness, plate width, etc. as keys. Furthermore, the target values ​​of the process parameters may be changed during rolling by manual intervention by an operator.

[0039] The setting calculator 23 has model formulas that represent the physical phenomena of each process, such as heating, rolling, cooling, and transportation, in the rolling line 1. The setting calculator 23 performs setting calculations using the model formulas that represent the physical phenomena of the processes so as to achieve the target values ​​(process conditions) of the various process parameters described above in actual operation. In the setting calculations, calculation of control target values ​​for various actuators and calculation of the state of the rolled material at each stage of the process (predicted state value of the metal material) are repeatedly performed.

[0040] The control target values ​​of the actuators include the roll gap of the rolling mills 6 and 14, the rolling speed, the conveying speed, the flow rates of the descaler and various sprays, the ON / OFF state of the valves on the run-out table, etc. The state of the rolled material at each stage of the process (predicted state values ​​of the metal material) includes the dimensions, shape, temperature, microstructure, etc.

[0041] The control controller 24 receives the setting calculation results from the setting computer 23 and controls various actuators so as to follow the control target values. In an actual hot rolling process, various sensors are installed throughout the rolling line 1 to monitor and collect actual values ​​of parameters that affect process control, such as temperature, shape, plate thickness, plate width, and rolling load.

[0042] These actual values ​​are used for process control, improving the accuracy of model formulas, and quality control. The target values ​​of the process parameters are compared with actual values ​​acquired by various sensors, or with calculated actual values ​​recalculated by the setting computer 23 from the actual values ​​and calculated values, and if the target values ​​of the process parameters are not achieved, the setting computer 23 performs the setting calculation again. Based on the results, various types of control, such as feedforward control, feedback control, and dynamic control, are performed.

[0043] Even when a process model equation accurately simulates physical phenomena, model prediction errors occur in reality. Therefore, engineers fine-tune the coefficients and constants for each term in the model equation to improve the prediction accuracy of the model equation. The adjustment terms are the coefficients and constants for each term in the model equation, and are managed in a database belonging to the setting computer 23 for each stratification using a stratification table classified by factors that are likely to cause model errors, such as steel grade, target plate thickness, target plate width, and target temperature. In addition to the start-up of operations, adjustment terms are mainly adjusted when rolling a new steel grade or a new combination of process parameters. The adjustment terms are sometimes adjusted by engineers based on experience or numerical analysis results, and in recent years, they have also been semi-automatically adjusted using statistical techniques such as neural networks. The learning terms are terms that are multiplied and added to the model equation to compensate for errors between the model output and the actual process output.

[0044] [Device for predicting material properties of rolled product] Figure 3 is a block diagram showing the functions of a device 25 for predicting material properties of a rolled product according to embodiment 1. The prediction device 25 predicts the mechanical properties of a product coil rolled in the rolling line 1 shown in Figure 1. The prediction device 25 includes an approximate model creation unit 26 and a material property prediction unit 27.

[0045] FIG. 4 is a diagram showing an example of the hardware configuration of the prediction device 25. Each function of the prediction device 25, which will be described later, can be realized by a processing circuit 250 shown in FIG. 4. This processing circuit 250 may be dedicated hardware 251. This processing circuit may include a processor 252 and a memory 253. This processing circuit may be partially formed as the dedicated hardware 251, and may further include the processor 252 and the memory 253. In the example of FIG. 4, part of the processing circuit 250 is formed as the dedicated hardware 251, and the processing circuit 250 also includes the processor 252 and the memory 253.

[0046] The processing circuitry 250 may be at least one dedicated hardware 251. In this case, the processing circuitry 250 may be, for example, a single circuit, a multiple circuit, a programmed processor, parallel programmed processors, an ASIC, an FPGA, or a combination thereof.

[0047] The processing circuit 250 may include at least one processor 252 and at least one memory 253. In this case, each function of the prediction device 25 is realized by software, firmware, or a combination of software and firmware. The software and firmware are written as programs and stored in the memory 253. The processor 252 realizes the functions of the approximate model creation unit 26 and the material property prediction unit 27 by reading and executing the programs stored in the memory 253. The processor 252 is also called a CPU (Central Processing Unit), central processing unit, processing device, arithmetic unit, microprocessor, microcomputer, or DSP. The memory 253 corresponds to, for example, non-volatile or volatile semiconductor memory such as RAM, ROM, flash memory, EPROM, or EEPROM. In this way, the processing circuit 250 can realize each function of the prediction device 25 by hardware, software, firmware, or a combination thereof.

[0048] The approximation model creation unit 26 creates an approximation model 39 that comprehensively and quickly predicts the mechanical properties of a group of hot rolled products manufactured on the rolling line 1, i.e., rolled products (product coils) of all steel types, dimensions, and operating conditions that can be manufactured on the rolling line 1. Comprehensive means that the approximation model creation unit 26 not only covers the mechanical properties of rolled products that have been manufactured on the rolling line 1, but also covers the mechanical properties of rolled products that may be manufactured on the rolling line 1 in the future. The approximation model creation unit 26 will be described in detail below.

[0049] 5 is a block diagram showing the configuration of the approximation model creation unit 26. The approximation model creation unit 26 includes a data set creation unit 28 and a model parameter determination unit 29. The data set creation unit 28 creates a data set used to create an approximation model 39. The model parameter determination unit 29 uses the data set created by the data set creation unit 28 to determine parameters that express the approximation model 39 by an optimization method or the like.

[0050] FIG. 6 is a diagram illustrating a dataset created by the dataset creation unit 28. The dataset 36 created by the dataset creation unit 28 is a collection of data pairs (pairs) consisting of explanatory variable 37 data and objective variable 38 data. Each pair is composed of explanatory variable 37 data corresponding to a portion of a certain operating condition and objective variable 38 data corresponding to the mechanical properties of the coil produced by the rolling line 1 under that operating condition. There are various operating conditions, but as the explanatory variables 37, operating conditions (hereinafter referred to as "first operating conditions") that are particularly correlated with the mechanical properties are selected, such as chemical composition, temperature conditions, processing conditions, time conditions, and speed conditions. The content of the chemical composition in the steel is adjusted to produce different steel grades. Chemical components that contribute to mechanical properties, such as C, Mn, Si, Nb, N, and Ti, are included in the explanatory variables 37. Like the chemical composition, the temperature conditions are adjusted to produce different steel grades, and contribute significantly to mechanical properties. The temperature history of a certain portion of the product coil is time-series data, but when provided as explanatory variables 37, one-dimensional data obtained by instantaneously cutting the time-series data at key points is provided. For example, one-dimensional data such as heating furnace exit temperature = 1200°C, rough rolling entry temperature = 1100°C, rough rolling exit temperature = 1000°C, finish rolling entry temperature = 950°C, finish rolling exit temperature = 900°C, and coiling temperature = 650°C are provided for each explanatory variable 37 to represent the main temperature history that affects the mechanical properties. Temperatures that can be actually measured with a thermometer or scan pyrometer during online product rolling and temperatures recalculated based on actual results by the setting calculator 23 are selected as explanatory variables 37, and are temperature conditions that affect the mechanical properties. Temperature conditions that may vary from product to product, such as the heating furnace exit temperature and the rough rolling entry temperature, are used as explanatory variables 37. Temperature conditions that vary in the rolling direction and width direction, such as the finish rolling exit temperature and coiling temperature, and temperature conditions that cause a temperature drop at the width ends, such as the rough rolling entry temperature and rough rolling exit temperature, are also used as explanatory variables 37. Furthermore, mechanical properties are engineered by adjusting the temperature path according to the cooling conditions in the run-out table 17.Specifically, in addition to the target temperatures of the finish delivery temperature and the coiling temperature, some of the cooling pattern conditions, such as the location of the water-cooled section, the water-cooling rate, the water-cooling time, and the air-cooling time, are set and controlled (normal cooling). Furthermore, the run-out table 17 may be divided into a first half and a second half, with an intermediate thermometer installed at the midpoint between them, and in addition to the target temperatures, the target temperatures of the intermediate thermometer and the cooling pattern conditions may be set and controlled for each of the first and second half of the run-out table 17 (step cooling). Therefore, the cooling pattern conditions, which will be described later, are also added to the explanatory variables. The processing conditions are operational conditions such as the reduction rate and strain rate of each stand. It is known that the reduction rate and strain rate affect the microstructure and, as a result, the final mechanical properties. These processing conditions are added to the explanatory variables as processing conditions for each stand. Alternatively, aggregate information, such as an average for the entire finish rolling process or an average for the second half stands of finish rolling, may be added to the explanatory variables. The time conditions, such as the reheating furnace time, which is the time the slab material is in the reheating furnace, and the rough-finishing time, which is the time from the end of rough rolling to the start of finish rolling, vary from product to product. Adding these conditions, which affect the mechanical properties, to the explanatory variables may improve the prediction performance of the variation in mechanical properties for each product. Furthermore, including speed conditions, such as the rolling speed and conveying speed at each location, which change from the leading end to the tail end in the rolling direction and affect the mechanical properties, as explanatory variables may improve the prediction performance of the variation in mechanical properties within a product. Mechanical properties that serve as the dependent variable include yield stress, tensile strength, and elongation. The dataset 36 comprehensively includes the operating conditions at all locations, including the leading end, tail end, and widthwise ends, of coils of all steel types and dimensions produced on the rolling line 1, as well as the mechanical properties when rolled under those operating conditions.

[0051] FIG. 7 is a schematic diagram for explaining an example of a cooling pattern of explanatory variables 37. FIG. 7 shows an example of a cooling pattern for normal cooling. In the example of FIG. 7, the upstream bank of the run-out table 17 is used for feedforward control using the initial cooling setting obtained by setting calculation and the actual value of the finishing temperature. The two most downstream banks are used for feedback control using the actual winding temperature, and the bank upstream of the bank used for feedback control is used for dynamic control to follow changes in the conveying speed. In order to express the cooling pattern as shown in FIG. 7 using explanatory variables, the explanatory variables are, for example, the water cooling section cooling rate, the water cooling section cooling rate of the front half, the water cooling section cooling rate of the rear half, the average cooling rate of the front half, the average cooling rate of the rear half, and the water cooling section cooling rate of the feedback bank. For example, the water cooling section cooling rate V cool is calculated using the following formula (1), which covers the temperature drop due to water cooling in banks excluding the dynamic control bank and feedback control bank.

[0052] where t W_s is the water cooling start time excluding the dynamic control bank and the feedback control bank, and t W_e is the water cooling end time excluding the dynamic control bank and the feedback control bank, and T W_s is the temperature at the start of water cooling, and T W_e is the temperature at the end of water cooling.

[0053] It is known that cooling in the first half and cooling in the second half of the run-out table 17 have different effects on mechanical properties. The cooling rate of the first half of the water-cooled section, the cooling rate of the second half of the water-cooled section, the average cooling rate of the first half, and the average cooling rate of the second half are calculated and used as explanatory variables. For example, the cooling rate V of the first half of the water-cooled section E_cool is the temperature drop due to water cooling in the bank between the finish outlet temperature and the intermediate temperature, and is calculated using the following formula (2).

[0054] where t E_s is the time when water cooling starts for the first half bank, and t E_e is the time when water cooling of the first bank ends, and T E_s is the temperature at the start of water cooling in the first half bank, and TE_e is the temperature at the end of water cooling in the first half bank.

[0055] The cooling rate of the water-cooled section in the latter half is also included in the explanatory variables by calculating it using the time and temperature information in the latter half. For example, the average cooling rate V E_ave is calculated using the following formula (3), which targets the temperature drop due to water cooling and air cooling in the bank between the finish outlet temperature and the intermediate temperature.

[0056] where t FDT is the time when the temperature reaches the final outlet temperature, and t MT is the time when the intermediate temperature is reached, and T FDT is the temperature at the outlet of the finished product, and T MT is the intermediate temperature.

[0057] The average cooling rate in the latter half of the roll is also included in the explanatory variables by calculating it using the time and temperature information for the latter half of the roll. Feedback control to control the coiling temperature can cause the cooling pattern by the feedback bank to differ in the rolling direction of the rolled material, which can affect the mechanical properties. To take such disturbances into account, the cooling rate in the water-cooled section of the feedback bank is added to the explanatory variables. Cooling rate V in the water-cooled section of the feedback bank FB_cool is calculated using the following equation (4) for the water-cooled part of the feedback bank.

[0058] where t FB_s is the feedback bank water cooling start time, and t FB_e is the feedback bank water cooling end time, and T FB_s is the feedback bank water cooling start temperature, and T FB_e is the feedback bank water cooling start temperature.

[0059] FIG. 8 is a diagram illustrating the configuration of the dataset creation unit 28 and the creation of a dataset. The dataset creation unit 28 includes a condition setting unit 30 and a material calculation unit 31. The condition setting unit 30 sets first operating conditions related to the explanatory variables 37 of the dataset 36. The condition setting unit 30 also sets, in the material calculation unit 31, operating conditions such as chemical components, processing history, and temperature history required for metallographic structure calculation using a metallurgical phenomenon model in the metallographic structure calculation unit 44 (hereinafter, the operating conditions related to the input of the metallographic phenomenon model used by the metallographic structure calculation unit 44 are referred to as "second operating conditions"). The first and second operating conditions may include the same data but may also include different data. For example, the metallurgical phenomenon model predicts changes in the microstructure over time. Therefore, the second operating conditions include data that are spatially and temporally continuous at each facility location and each time, such as processing history and temperature history. The material calculation unit 31 calculates metallurgical phenomena and mechanical properties based on the set second operating conditions. Furthermore, the material calculation unit 31 sets data for the objective variable 38 of the data set 36 .

[0060] 9 is a diagram for explaining the setting of the material calculation unit 31 and explanatory variables 37 by the condition setting unit 30. The condition setting unit 30 is composed of, for example, an operating condition extraction unit 40, an operating condition creation unit 41, a material calculation input setting unit 42, an explanatory variable setting unit 43, and the like.

[0061] The operating condition extraction unit 40 extracts, for example, first operating conditions related to the above-mentioned explanatory variables 37 such as chemical components, processing history, and temperature history, and second operating conditions to be used in the material calculation unit 31, from virtual rolling data of a virtual coil virtually rolled by an offline setting computer.

[0062] The offline setting computer is, for example, a device such as that described in Patent Document 1, and can simulate actual operations by synchronizing process parameters with the online setting computer. This allows a variety of operating conditions to be created by offline setting calculations without imposing a load on actual operations. For example, even for products and rolling conditions that have not been manufactured in actual operations, it is possible to simulate operations that satisfy various machine constraints.

[0063] The operating conditions and explanatory variables 37 used in the material calculation unit 31 may include operating conditions and explanatory variables 37 extracted from actual rolling data of a product coil that has actually been rolled. Alternatively, operating conditions such as processing history and temperature history, including strain at each part in the product, may be created using analysis data that is more detailed than the set calculation, such as the results of finite element analysis.

[0064] The operating conditions to which the approximation model 39 can be applied should include disturbances that are not measured or calculated in actual operation. Examples of disturbances include variations in chemical composition, temperature variations during reheating in a heating furnace, a temperature drop in the rolled material due to unexpected oscillation before finish rolling, and an inability to ensure the cooling rate due to a spray malfunction. The disturbances are simulated using the offline setting computer. Alternatively, the disturbances may be directly added to the processing history or temperature history of the actual rolling data or virtual rolling data.

[0065] The operating condition creating unit 41 intentionally modifies a part of the rolling data created by duplicating the actual rolling data and the virtual rolling data to simulate a disturbance.

[0066] The material calculation input setting unit 42 passes to the material calculation unit 31 the operating conditions extracted by the operating condition extraction unit 40, or the operating conditions extracted by the operating condition extraction unit 40 with changes made by the operating condition creation unit 41 corresponding to disturbances, or the operating conditions created by the operating condition creation unit 41 from external data such as finite element analysis results without going through the operating condition extraction unit 40.

[0067] The explanatory variable setting unit 43 extracts data corresponding to the explanatory variables 37 from the operating conditions extracted by the operating condition extraction unit 40, or from the operating conditions extracted by the operating condition extraction unit 40 to which the operating condition creation unit 41 has made changes corresponding to disturbances, or from the operating conditions created by the operating condition creation unit 41 without going through the operating condition extraction unit 40, and adds the extracted data to the dataset 36 as the explanatory variables 37.

[0068] 10 is a diagram for explaining the metallurgical phenomenon calculation and the setting of the objective variable 38 by the material calculation unit 31. The material calculation unit 31 is made up of a metal structure calculation unit 44, a material property calculation unit 45, and an objective variable setting unit 46.

[0069] The metallographic structure calculation unit 44 calculates the metallographic structure using a metallurgical phenomenon model that mathematically expresses metallurgical phenomena, using rolling data such as chemical composition, processing history, and temperature history set in the material calculation input setting unit 42 of the condition setting unit 11. The metallographic characteristics to be calculated include the volume fractions of ferrite, pearlite, bainite, and martensite, as well as the grain sizes of ferrite and austenite. Various metallurgical phenomenon models have been proposed, each consisting of a group of mathematical expressions representing static recovery, static recrystallization, dynamic recovery, dynamic recrystallization, grain growth, and transformation. An example of this model is published on pages 198-229 of "Plastic Processing Technology Series 7: Plate Rolling" (Corona Publishing). Using this model, it is possible to calculate the volume fractions of ferrite, pearlite, bainite, martensite, and the like, as well as the austenite and ferrite grain sizes. To handle micro-order changes in the structure, the calculation steps and calculation domain must be subdivided, resulting in a heavy computational load. The dataset 36 must contain a huge number of pairs of explanatory variables and target variables, representing thousands to tens of thousands of coils. In order to operate the metal structure calculation unit 44 under a variety of operating conditions, ranging from several thousand to several tens of thousands of cases, calculations are performed in an offline environment that does not affect the online calculations of actual operations.

[0070] The material property calculation unit 45 calculates mechanical properties based on the chemical compositions and other data contained in the rolling data and the metallographic structure calculation values ​​obtained from the metallographic structure calculation unit 44. The mechanical properties to be measured include yield stress, tensile strength, elongation, and the like. Metallurgical phenomena models that mathematically represent metallurgical phenomena are often automatically trained on an ongoing basis using actual mechanical property values ​​obtained from mechanical property measurement tests, such as tensile tests, conducted on a portion of the product coil (e.g., the method described in Patent Document 4). However, the measured mechanical properties are measured on test specimens cut from a portion of the coil, such as the head or tail end of the coil. Therefore, the changes in mechanical properties due to differences in operating conditions, such as the temperature in the rolling direction and the width direction, are generally not measured and are not reflected in the training of the metallurgical phenomena model. Under operating conditions where such measured mechanical properties are obtained, the mechanical property prediction results of the metallurgical phenomena model, including the training, become idiosyncratic values, resulting in a non-smooth change in the mechanical property prediction results relative to changes in the operating conditions. The material property calculation unit 45 uses a metallurgical phenomenon model that does not involve learning, rather than a metallurgical phenomenon model that has been automatically learned in actual operation.

[0071] The objective variable setting unit 46 extracts data corresponding to the objective variable 20 from the mechanical properties created by the material property calculation unit 45, and adds the extracted data to the data set 36 as the objective variable 20 that forms a pair with the explanatory variable 37 created from the operating conditions used by the material calculation unit 12.

[0072] The model parameter determination unit 29 uses the data set 36 created by the data set creation unit 28 to determine model parameters that constitute an approximation model 39 that simply simulates the mechanical property prediction calculation that has undergone the metal structure prediction with a large load in the material calculation unit 12.

[0073] FIG. 11 is a schematic diagram showing an example of an approximation model. A machine learning model can be used as the approximation model 39. In the example shown in FIG. 11, the machine learning model is constructed by a forward propagation neural network consisting of an input layer, an intermediate layer, and an output layer. Explanatory variables 37, which are operating conditions, are input to the input layer. A response variable 20, which is a mechanical property, is output from the output layer.

[0074] In an example of constructing an approximation model using a feedforward neural network, the model parameter determination unit 29 determines the hyperparameters and parameters of the feedforward neural network. Hyperparameters include the network configuration (number of hidden layers, number of units), the type of activation function, etc. The hyperparameters are determined empirically or by trial and error. Alternatively, they are determined using commonly used methods such as grid search and Bayesian optimization. The parameters are the weight coefficients and biases of each unit that represent the function of the feedforward neural network. Although the function that accurately predicts mechanical properties under all operating conditions is unknown, given a large number of input-output pairs of the function, a function that closely reproduces these input-output pairs can be created by adjusting the parameters. The input-output pairs are called training data. Parameters are selected so that when the training data input (explanatory variables) is applied to the function, the output of the neural network is as close as possible to the training data output (objective variable). The explanatory variables 37 of the dataset 36 created by the dataset creation unit 28 are used as inputs, and the objective variable 20 is used as outputs. The parameters are adjusted using the pairs of explanatory variables 37 and objective variable 20. For training, a portion of the pairs in the dataset, for example, approximately 70% of the total, is used as training data. An error function, such as squared error, is used as a measure of the reproducibility of the function represented by the neural network. Neural networks are trained by solving the problem of minimizing the error function. For training a forward propagation neural network, gradient descent and its improved method, Adam, are known as optimization methods for solving the problem of minimizing the error function. For example, the model parameter determination unit 29 trains the forward propagation neural network using Adam. After confirming the model generalization performance using the holdout method or cross-validation, the parameters may be adjusted again as necessary after implementing measures commonly used in building machine learning models, such as changing hyperparameters, increasing the number of data pairs in the dataset, or reviewing data preprocessing.Furthermore, after constructing the machine learning model according to the above procedure, if new pairs of explanatory variables 37 and objective variables 20 are added to the data set as needed, sequential learning may be performed using a gradient descent method or the like. This is the case when online rolling data is obtained as needed, or when offline setting calculations are performed as needed. Furthermore, the data set may be recreated using the latest data at the time of turnaround maintenance, and the approximation model may be reconstructed and replaced. It is also possible to change the type of explanatory variables depending on the type of mechanical property to be predicted. It is preferable to create a different approximation model for each mechanical property.

[0075] The material property prediction unit 27 predicts online the mechanical properties of each three-dimensional mesh area of ​​the product coil actually manufactured on the rolling line 1 using an approximate model 39 that has been created in advance offline by the approximate model creation unit 26.

[0076] FIG. 12 is a block diagram showing the configuration of the material property prediction unit 27. As shown in FIG. 12, the material property prediction unit 27 includes a rolling data collection unit 32, a model input creation unit 33, an approximation model calculation unit 34, and a material property output unit 35. The rolling data collection unit 32 collects rolling data for each portion of the coil from the heating furnace to winding. The model input creation unit 33 creates input data for an approximation model 39 from the rolling data collected by the rolling data collection unit 32. The approximation model calculation unit 34 calculates the mechanical properties of each three-dimensional mesh-like region of the product coil using the input data created by the model input creation unit 33 and the approximation model 39. The material property output unit 35 outputs the mechanical properties calculated by the approximation model calculation unit 34 together with information indicating the position within the product.

[0077] The rolling data collection unit 13 collects rolling data such as the chemical composition, processing history, and temperature history of the product coil calculated by the setting computer 23. The rolling data is a predicted value before rolling, a predicted value and a recalculated actual value during rolling, and a recalculated actual value after rolling, and the accuracy of the information varies depending on the timing of acquisition. When predicting mechanical properties for the purpose of product quality control, it is preferable to collect data when the recalculated actual values ​​after winding is completed by the down coiler 19 are available.

[0078] The model input creation unit 33 extracts data corresponding to the explanatory variables 37 of each three-dimensional mesh-like zone of the product coil from the rolling data of the product coil collected by the rolling data collection unit 13, as input data for the approximation model 39. If the rolling data collected by the rolling data collection unit 13 is insufficient for creating the explanatory variables 37 of each three-dimensional mesh-like zone, the explanatory variables 37 are created by supplementing the missing data. For example, in recent years, the temperature distributions in the thickness and width directions of the rolled material are sometimes calculated based on actual measurements taken with a thermometer and model predictions, but due to the calculation load, the setting computer 23 may intentionally limit the calculation range, for example, by calculating the width direction temperature distribution up to the outlet side of an edge heater required for control and stopping the calculation downstream of the edge heater. In such cases, the rolling data collection unit 13 uses downstream data, such as the actual width direction temperature distribution measured by a multi-gauge at the exit of the finishing mill, the model-calculated width direction temperature distribution at the exit of the edge heater or the actual width direction temperature distribution measured by a scan pyrometer at the exit of the edge heater, and the temperature history calculation results for the width direction center portion at the exit of the finishing mill to create explanatory variable 37 data related to the temperature of each mesh in the width direction, such as the temperature at the entry side of the finish rolling, using an interpolation formula such as linear interpolation. Alternatively, the width direction temperature distribution may be calculated using an offline setting calculator. Furthermore, some of the processing history, such as the strain rate during rolling in the thickness direction, is generally not calculated in the width direction. This is because, in thickness direction rolling, the contact length between the roll and the deformation region of the material is significantly shorter than the width of the material, so the material does not move much in the width direction, and the thickness reduction is mainly due to elongation in the rolling direction. In this case, the data calculated at the width direction center portion can be replicated in the width direction and used as explanatory variable 37. In addition, for data such as chemical components for which only one value can be obtained for a product coil, the same value is used for all meshes.

[0079] The approximate model calculation unit 34 inputs the input data created by the model input creation unit 33 into the approximate model 39 created in advance by the approximate model creation unit 26, and calculates the mechanical properties of each three-dimensional mesh area of ​​the product coil.

[0080] FIG. 13 is a schematic diagram of an example of a mesh area for which the approximation model calculation unit 34 calculates material properties. For simplicity, FIG. 13 shows only a portion of the total length and one half of the strip width direction (work side or drive side). However, the approximation model calculation unit 34 calculates the entire length and the entire width. The calculation interval set in the rolling direction may vary depending on the product thickness. For example, rolling data is created at 2-m intervals, and mechanical properties are also predicted at similar intervals. In the thickness direction, mechanical properties are predicted at three points, for example, the top surface, the thickness center, and the bottom surface. In the width direction, mechanical properties are predicted at a total of five points, for example, a point 40 mm from the width end (work side and drive side), a midpoint between the width end and the width center (1 / 4 strip width point) (work side and drive side), and the strip width center.

[0081] The material property output unit 35 outputs the mechanical properties of each zone calculated by the approximation model calculation unit 34, information representing the position of each zone within the rolled product, and information closely related to variations in mechanical properties. The positions of each zone within the rolled product are the distance from the leading edge in the rolling direction, the distance from the tail edge, the widthwise position, and the thicknesswise position. The information relevant to the mechanical properties is information closely related to variations in mechanical properties, such as the finish delivery temperature, the coiling temperature, and the cooling rate. The output destination is a storage device such as a database and a visualization device such as an HMI. When output to a visualization device, the output data is visualized in a graph or numerically displayed in a table. These values ​​are used as reference values ​​when considering whether or not to conduct a tensile test, determining the product grade, the cut-off length at the dividing line, and process parameters for rolled products of the same lot. Because the mechanical properties of each zone, information representing the position within the rolled product, and information closely related to variations in mechanical properties can be obtained quickly, they can be reflected in the process parameters for the next material or for materials of the same lot.

[0082] FIG. 14 is a schematic diagram showing an example of the output from the material property output unit 35. The horizontal axis represents the rolling direction position (distance from the front end), and the vertical axis represents the changes in tensile strength and coiling temperature. The positions in the thickness direction and width direction can be selected arbitrarily; in this example, the center of the thickness and the work side 1 / 4 point in the width direction are selected. When a tensile strength lower limit is given, information such as the position of the rolled material where the tensile strength falls below the lower limit and its relationship with the coiling temperature can be obtained. Changing the horizontal axis to the thickness direction or width direction makes it possible to visualize the distribution of mechanical properties in each direction. Another output method is to visualize the distribution of mechanical properties for the entire product using a three-dimensional mesh-like heat map. Information visualized in the table includes, for example, information on the position where mechanical properties are not achieved (distance from the front and rear ends), basic statistics such as the average value and standard deviation of the mechanical properties, and the correlation coefficient between the mechanical properties and the explanatory variables 37.

[0083] As described above, according to this embodiment, a comprehensive data set is created in advance, even for rolling conditions (temperature conditions, processing conditions, time / speed conditions) that have not been implemented in an actual rolling process or for which little experience has been gained. By creating an approximation model offline using a data set created so as to encompass all rolling conditions, including those that have not been implemented or for which little experience has been gained, it is possible to create an approximation model that can be used for the entire group of rolled products and has high approximation accuracy. By using the approximation model created in this way, it is possible to reduce the computational load required for online prediction of material properties in each zone of a three-dimensional mesh of all rolled products. Furthermore, it is possible to calculate and monitor material properties in all zones (portions) of a rolled product. This makes it possible to reflect the results of the rolled product in operational changes for the next rolled product or for rolled products in the same lot.

[0084] Second Embodiment Next, a second embodiment of the present invention will be described with reference to Figures 15 and 16. The description will focus on the differences from the first embodiment described above, and the same or corresponding parts will be denoted by the same reference numerals and description thereof will be omitted.

[0085] 15 is a block diagram showing the configuration of a material property prediction unit 27 according to the second embodiment. The mechanical property prediction unit 27 as a material property prediction unit includes a material property correction unit 47 that corrects the mechanical property calculated by the approximation model calculation unit 34 using the approximation model 39 (hereinafter referred to as the "first mechanical property") using the mechanical property calculated by a metallurgical phenomenon model that mathematically expresses a metallurgical phenomenon (hereinafter referred to as the "second mechanical property"). The metallurgical phenomenon model that mathematically expresses a metallurgical phenomenon and predicts the second mechanical property is model-trained using actual mechanical property values ​​obtained in tensile tests of past rolled products (product coils). The second mechanical property is collected as part of rolling data by the rolling data collection unit 13 from the online setting computer 23.

[0086] The mechanical properties of a portion of a rolled product, for example, at a representative point of the head, center, or tail end in the rolling direction, and at the center of each width direction, may be calculated using a metallurgical phenomenon model that mathematically represents metallurgical phenomena, using an online setting computer or a computer connected to the online setting computer. Furthermore, the metallurgical phenomenon model that mathematically represents metallurgical phenomena is trained using actual values ​​of mechanical properties obtained from mechanical property measurement tests, such as tensile tests and microstructural observations, performed on some product coils, or is adjusted to match the actual mechanical properties (e.g., the method described in Patent Document 4). The prediction accuracy of the second mechanical property thus trained or adjusted can be expected to be good.

[0087] An example of a method for correcting the first mechanical property by the material characteristic correction unit 47 will be described below. It is assumed that the second mechanical property is obtained in area j among the mesh areas for which the approximate model calculation unit 34 calculates the first mechanical property. In the following, an example will be described in which the mechanical property is tensile strength.

[0088] First, in the mesh area j where the second mechanical property is obtained, the difference between the second mechanical property and the first mechanical property is calculated using the following formula (5): ΔTS(j)=TS MM (j)-TS ML (j) ... (5) where j is an index indicating the mesh area where the second mechanical property was obtained, and TS MM(j) is the secondary mechanical property included in the actual rolling data, and TS ML (j) is the first mechanical property calculated by the approximate model calculation unit 34, and ΔTS(j) is the difference between the second mechanical property and the first mechanical property.

[0089] 15 is a schematic diagram showing a correction method in the material property correction unit 47 in the second embodiment of the present invention. For example, when the second mechanical properties are obtained at the head representative point, the center representative point, the tail representative point, and the center of the width direction of each of the rolling direction representative points, the difference ΔTS(i) between the second mechanical property and the first mechanical property assumed in another mesh area i is calculated by linear interpolation or the like using ΔTS(j) at the above three points.

[0090] Subsequently, the mesh area i is corrected using the difference ΔTS(i) between the second mechanical property and the first mechanical property using the following formula (6): TS comp (i)=TS ML (i) + α·ΔTS(i) (6) where i is an index indicating the mesh area for which the tensile strength is to be corrected, and TS ML (i) is the tensile strength calculated by the approximation model calculation unit 34 using the approximation model 39, α is a correction adjustment coefficient (=0 to 1), and TS comp (i) is the corrected tensile strength.

[0091] As described above, according to this embodiment, since the material property correction unit 47 is provided, it is possible to improve the prediction accuracy of the mechanical properties of all parts of the rolled product. Moreover, there is no influence on the online calculations of the actual rolling operation. Note that although an example of predicting the mechanical properties of the material properties has been described above, the same applies to the electromagnetic properties.

[0092] Although the embodiments of the present invention have been described above, the present invention is not limited to the above-described embodiments and can be implemented in various modifications without departing from the spirit of the present invention. When the numbers, quantities, amounts, ranges, etc. of each element are mentioned in the above-described embodiments, the present invention is not limited to the mentioned numbers unless otherwise specified or clearly specified in principle. Furthermore, the structures, etc. described in the above-described embodiments are not necessarily essential to the present invention unless otherwise specified or clearly specified in principle.

[0093] In the above embodiment, the case where mechanical properties are predicted as material properties has been described as an example, but the same applies to the case where electromagnetic properties are predicted.

[0094] 1... Rolling line, 25... Rolled product material property prediction device, 26... Approximation model creation unit, 27... Material property prediction unit, 28... Data set creation unit, 29... Model parameter determination unit, 30... Condition setting unit, 31... Material calculation unit, 36... Data set, 37... Explanatory variable, 38... Objective variable, 47... Material property correction unit

Claims

1. A device for predicting material properties of a rolled product to be manufactured on a rolling line, comprising: an approximation model creating unit that creates an approximation model offline to comprehensively predict material properties of not only a group of rolled products actually produced on the rolling line but also a group of virtual rolled products that have not actually been produced on the rolling line but may be produced in the future; a material property prediction unit that predicts online material properties in each three-dimensional mesh-like area of ​​a rolled product produced in the rolling line by using the approximation model created by the approximation model creation unit, The approximation model creation unit is a condition setting unit that sets rolling conditions for the rolled product group, and a material calculation unit that calculates metallurgical phenomena and material properties under the rolling conditions, and a data set creation unit that creates a data set used to create the approximation model; a model parameter determination unit that determines parameters that represent the approximation model using the data set; The data set creation unit uses the rolling conditions extracted from rolling data of the rolled products actually manufactured on the rolling line and the virtual rolling conditions extracted from virtual rolling data of the virtual rolled products virtually rolled offline as explanatory variables, and creates the data set using the material properties calculated by the material calculation unit as objective variables.

2. The material property prediction unit is a rolling data collection unit that collects online rolling data obtained when a rolled product is manufactured in the rolling line; a model input creation unit that creates input data for the approximation model online from the rolling data collected by the rolling data collection unit; an approximation model calculation unit that inputs the input data created by the model input creation unit into the approximation model to online calculate material properties of each region of a three-dimensional mesh of a product coil; 2. The material property prediction device for a rolled product as described in claim 1, further comprising a material property output unit that outputs the material properties of each zone calculated by the approximation model calculation unit, information representing the position of each zone within the rolled product, and information related to the material properties.

3. The device for predicting material properties of a rolled product according to claim 1 , wherein the approximation model is a machine learning model.

4. 4. The material property prediction device for a rolled product according to claim 1, wherein the material property prediction unit is provided with a material property correction unit that corrects the material properties calculated by the approximation model calculation unit using the approximation model using material property results calculated using a metallurgical phenomenon model that mathematically represents a metallurgical phenomenon.

Images